me 599/ee 546 - faculty.washington.edu
TRANSCRIPT
ME 599/EE 546: Biology-inspired
roboticsLecture 1
Sawyer B. Fuller
Goals: • Describe the need for “biology-inspired robotics” • Describe how this course works
biology-inspired robot control• MW 10:30-11:50 in MEB259
• Instructor: Prof. Sawyer B. Fuller ([email protected])
• Office hours: Wednesdays 1:30-2:20 in MEB 321
• Prereq’s: an undergraduate degree in ME, EE, or Aero
• Website: http://faculty.washington.edu/minster/bio_inspired_robotics/
• lecture slides, papers, and other materials will be posted there
• submit all coursework on Canvas
2
Robot (noun)a machine capable of carrying out a complex series of
actions automatically
UW Classes that take a more classical approach to robotics:
• Mechatronics:
• ME581: Digital control systems (spring)
• Dynamics and control:
• EE 543/544: Kinematics and dynamics of robot arms/manipulators
• ME583: Nonlinear control
• Perception and planning:
• EE 576: computer vision and robotics (spring)
• CSE 571 Probabilistic Robotics: Perception, localization, mapping (fall)
• AMATH / CSE 579: Intelligent control through learning and optimization (Emo Todorov)
• ME599: Advanced Robotics: Perception and multi-robot control (fall. Ashis Banerjee)
• Also:
• CSE 590: Robotics colloquium seminar (weekly speakers)
• BI 427: Animal biomechanics (fall, Tom Daniel)
4
“biology-inspired robotics”
5
current state of the art
6
• very power hungry (20x human of same weight) • only in controlled environments
Honda’s Asimo
current state of the art
7
Google’s self-driving car
• power hungry - requires a bank of computers • only in controlled environments
application areas where current robotics is still outperformed by nature
8
tiny robots (minimal computation,
limited sensing)
complex environments and robots (models are inadequate
and behavior must be learned)
agile robots (limited time to compute)
soft robots (nonlinear
stress/strain curve hard to model)
animal locomotion
9
• rich behavioral repertoire
animal locomotion
10
Animal locomotion
11
• aggressive, dynamic motions
Animal locomotion
12
• complex environments • minimal energy expenditure on computation
Aerial autonomy at insect scale
(images to scale)
Autonomous Insect Robotics Laboratory Est. 2015
Dr. Sawyer B. Fuller 15
Ma, Chirarattananon, Fuller, and Wood, Science 2013
Dr. Sawyer B. Fuller 16
RoboBee flight control
Dr. Sawyer B. Fuller17
previous work
• external power • external sensing • external computation
• improved capabilities • onboard sensing • onboard computing • onboard power
current research
Dr. Sawyer B. Fuller18
355 nm laser micromachining
assembly
fold, add
piezo & wing
Dr. Sawyer B. Fuller
piezo actuation
20
wing
flexure joints
V+
Vs
bimorph actuator
Dr. Sawyer B. Fuller21compound eyes
gyroscopic halteres
(angular velocity)
ocelli(direction of sun/sky)
antennae (wind, smell,
sound, gravity)
flapping wings
current research
22
vision (tiny camera)
odor (using moth antenna) onboard power
biology-inspired robotics
23
this course focuses on two areas where biology excels
1. mechanical intelligence
2. adaptation through evolution and learning
24
mechanical intelligence
• system is stable without active feedback
25
this fish is dead!
Liao, 2004
robot mechanical intelligence
26collins 2001
walks with no feedback and very little power
example of adaptation: reflexive/model-free control
• minimal internal representation
• cascaded behaviors
27
how mud wasps build nests
Smith, 1978
example of model-free control
28bongard 2011
a gait learned by a neural network
how this course works• This is a research-oriented course
• Three parts to the coursework:
1. Reviewing and discussing 1-2 assigned papers per class session
2. Presenting 1-2 papers to the class
• assigned based on a lottery and your preferences
3. Final project
• this year: you’re the funding agency
29
This year’s final project• You’re the funding agency!
• each team/individual submits a research proposal at the end of the quarter
• format: 3 pages, NDSEG graduate fellowship format
• includes preliminary work you did in this course
• show a “proof-of-concept” initial work in some aspect of biology-inspired robotics, probably in simulation
• can be used to for your actual application
• There will be a peer vote for best proposals
• criteria: quality of preliminary results, future promise
• top 3 proposals split the money
30
Wednesday• due by Tuesday 9pm: Review of paper 0
• because of the short time available, only complete part 1 (synopsis) of the four elements of the review.
31
next week• paper 1 review due Sunday
• paper 2 review due Tuesday